The Agent Skills Directory

[SAFE]: The skill processes user writing samples locally using a Node.js script. Analysis of scripts/voice.js confirms that all file operations are scoped to paths provided by the user or within the working directory. It does not perform any direct network exfiltration or credential harvesting.
[COMMAND_EXECUTION]: The skill uses a Bash tool to execute scripts/voice.js. This is the intended primary purpose of the skill to perform bulk ingestion and compilation of writing profiles. The script includes safety checks to ensure the --account flag is regex-validated and prevents unauthorized file system traversal during automatic file naming.
[DATA_EXPOSURE_&_EXFILTRATION]: The skill defines a clear boundary between 'Judgment' (local) and 'Write' (hub). It reads from MCP resources (postey://posts, etc.) to build a corpus but explicitly instructs the agent to never commit, log, or send the resulting profile anywhere except the user's local memory or files.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external content (past posts). It mitigates injection risks by using deterministic detectors (scripts/voiceFeatures.js) that measure countable features (sentence length, emoji rate) rather than interpreting or executing the content of the posts. This ensures that even if a post contains malicious instructions, the script treats it as raw text for statistical analysis.

postey-voice